📄 calmanfp.asm
字号:
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; calmanfp.asm - floating point version of the calcmand.asm file
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; The following code was adapted from a little program called "Mandelbrot
; Sets by Wesley Loewer" which had a very limited distribution (my
; Algebra II class). It didn't have any of the fancy integer math, but it
; did run floating point stuff pretty fast.
;
; The code was originally optimized for a 287 ('cuz that's what I've got)
; and for a large maxit (ie: use of generous overhead outside the loop to get
; slightly faster code inside the loop), which is generally the case when
; Fractint chooses to use floating point math. This code also has the
; advantage that once the initial parameters are loaded into the fpu
; register, no transfers of fp values to/from memory are needed except to
; check periodicity and to show orbits and the like. Thus, values keep all
; the significant digits of the full 10 byte real number format internal to
; the fpu. Intermediate results are not rounded to the normal IEEE 8 byte
; format (double) at any time.
;
; The non fpu specific stuff, such as periodicity checking and orbits,
; was adapted from CALCFRAC.C and CALCMAND.ASM.
;
; This file must be assembled with floating point emulation turned on. I
; suppose there could be some compiler differences in the emulation
; libraries, but this code has been successfully tested with the MSQC 2.51
; and MSC 5.1 emulation libraries.
;
; Wes Loewer
;
; and now for some REAL fractal calculations...
; (get it, real, floating point..., never mind)
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; required for compatibility if Turbo ASM
IFDEF ??version
MASM51
QUIRKS
ENDIF
.8086
.8087
.MODEL medium,c
; external functions
EXTRN keypressed:FAR ; this routine is in 'general.asm'
EXTRN getakey:FAR ; this routine is in 'general.asm'
EXTRN plot_orbit:FAR ; this routine is in 'fracsubr.c'
EXTRN scrub_orbit:FAR ; this routine is in 'fracsubr.c'
; external data
EXTRN init:WORD ; declared as type complex
EXTRN parm:WORD ; declared as type complex
EXTRN new:WORD ; declared as type complex
EXTRN maxit:WORD
EXTRN inside:WORD
EXTRN outside:WORD
EXTRN fpu:WORD ; fpu type: 87, 287, or 387
EXTRN rqlim:QWORD ; bailout (I never did figure out
; what "rqlim" stands for. -Wes)
EXTRN color:WORD
EXTRN oldcolor:WORD
EXTRN realcolor:WORD
EXTRN periodicitycheck:WORD
EXTRN reset_periodicity:WORD
EXTRN closenuff:QWORD
EXTRN fractype:WORD ; Mandelbrot or Julia
EXTRN kbdcount:WORD ; keyboard counter
EXTRN dotmode:WORD
EXTRN show_orbit:WORD ; "show-orbit" flag
EXTRN orbit_ptr:WORD ; "orbit pointer" flag
EXTRN potflag:WORD ; potential flag
EXTRN magnitude:QWORD ; when using potential
JULIAFP EQU 6 ; from FRACTYPE.H
MANDELFP EQU 4
GREEN EQU 2 ; near y-axis
YELLOW EQU 6 ; near x-axis
initx EQU <qword ptr init> ; just to make life easier
inity EQU <qword ptr init+8>
parmx EQU <qword ptr parm>
parmy EQU <qword ptr parm+8>
newx EQU <qword ptr new>
newy EQU <qword ptr new+8>
; Apparently, these might be needed for TC++ overlays. I don't know if
; these are really needed here since I am not familiar with TC++. -Wes
FRAME MACRO regs
push bp
mov bp, sp
IRP reg, <regs>
push reg
ENDM
ENDM
UNFRAME MACRO regs
IRP reg, <regs>
pop reg
ENDM
pop bp
ENDM
.DATA
align 2
savedx DQ ?
savedy DQ ?
orbit_real DQ ?
orbit_imag DQ ?
close DD 0.01
round_down_half DD 0.5
tmp_word DW ?
inside_color DW ?
periodicity_color DW ?
;savedand DW ?
;savedincr DW ?
savedand EQU SI ; this doesn't save much time or
savedincr EQU DI ; space, but it doesn't hurt either
.CODE
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; This routine is called once per image.
; Put things here that won't change from one pixel to the next.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
PUBLIC calcmandfpasmstart
calcmandfpasmstart PROC
; not sure if needed here
FRAME <di,si> ; std frame, for TC++ overlays
mov ax,inside
cmp ax,0 ; if (inside color == maxiter)
jnl non_neg_inside
mov ax,maxit ; use maxit as inside_color
non_neg_inside: ; else
mov inside_color,ax ; use inside as inside_color
cmp periodicitycheck,0 ; if periodicitycheck < 0
jnl non_neg_periodicitycheck
mov ax,7 ; use color 7 (default white)
non_neg_periodicitycheck: ; else
mov periodicity_color,ax ; use inside_color still in ax
mov oldcolor,0 ; no periodicity checking on 1st pixel
sub ax,ax ; ax=0
UNFRAME <si,di> ; pop stack frame
ret
calcmandfpasmstart ENDP
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; floating point version of calcmandasm
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
PUBLIC calcmandfpasm
calcmandfpasm PROC
FRAME <di,si> ; std frame, for TC++ overlays
; initialization stuff
sub ax,ax ; clear ax
cmp periodicitycheck,ax ; periodicity checking?
je initoldcolor ; no, set oldcolor 0 to disable it
cmp inside,-59 ; zmag?
je initoldcolor ; set oldcolor to 0
cmp reset_periodicity,ax ; periodicity reset?
je short initparms ; no, inherit oldcolor from prior invocation
mov ax,maxit ; yup. reset oldcolor to maxit-250
sub ax,250 ; (avoids slowness at high maxits)
initoldcolor:
mov oldcolor,ax ; reset oldcolor
initparms:
sub ax,ax ; clear ax
mov word ptr savedx,ax ; savedx = 0.0
mov word ptr savedx+2,ax ; needed since savedx is a QWORD
mov word ptr savedx+4,ax
mov word ptr savedx+6,ax
mov word ptr savedy,ax ; savedy = 0.0
mov word ptr savedy+2,ax ; needed since savedy is a QWORD
mov word ptr savedy+4,ax
mov word ptr savedy+6,ax
inc ax ; ax = 1
mov savedand,ax ; savedand = 1
mov savedincr,ax ; savedincr = 1
mov orbit_ptr,0 ; clear orbits
dec kbdcount ; decrement the keyboard counter
jns short nokey ; skip keyboard test if still positive
mov kbdcount,10 ; stuff in a low kbd count
cmp show_orbit,0 ; are we showing orbits?
jne quickkbd ; yup. leave it that way.
;this may need to be adjusted, I'm guessing at the "appropriate" values -Wes
mov kbdcount,5000 ; else, stuff an appropriate count val
cmp fpu,387 ; ("appropriate" to the FPU)
je short kbddiskadj ; ...
mov kbdcount,3000 ; ...
cmp fpu,287 ; ...
je short kbddiskadj ; ...
mov kbdcount,1000 ; ...
cmp fpu,87 ; ...
je short kbddiskadj ; ...
mov kbdcount,500 ; emulation
kbddiskadj:
cmp dotmode,11 ; disk video?
jne quickkbd ; no, leave as is
mov cl,2 ; yes, reduce count
shr kbdcount,cl ; ...
quickkbd:
call far ptr keypressed ; has a key been pressed?
cmp ax,0 ; ...
je nokey ; nope. proceed
mov kbdcount,0 ; make sure it goes negative again
cmp ax,'o' ; orbit toggle hit?
je orbitkey ; yup. show orbits
cmp ax,'O' ; orbit toggle hit?
jne keyhit ; nope. normal key.
orbitkey:
call far ptr getakey ; read the key for real
mov ax,1 ; reset orbittoggle = 1 - orbittoggle
sub ax,show_orbit ; ...
mov show_orbit,ax ; ...
jmp short nokey ; pretend no key was hit
keyhit:
mov ax,-1 ; return with -1
mov color,ax ; set color to -1
UNFRAME <si,di> ; pop stack frame
ret ; bail out!
nokey:
; OK, here's the heart of the floating point code.
; In my original program, the bailout value was loaded once per image and
; was left on the floating point stack after each pixel, and finally popped
; off the stack when the fractal was finished. A lot of overhead for very
; little gain in my opinion, so I changed it so that it loads and unloads
; per pixel. -Wes
fld rqlim ; everything needs bailout first
mov cx,maxit ; using cx as loop counter
cmp fpu,387 ; jump to fpu specific code
je start_387 ; 387, slight efficiency tweeking
cmp fpu,287 ;
je to_start_287 ; 287 (original version)
jmp start_87 ; else must be 87/emulation
to_start_287:
jmp start_287 ; needs a long jump here
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; _387 code is just like _287 code except that it uses an FADD instead
; of an FSCALE per orbit and also saves an FLD1 per pixel.
;
; You could use .386/.387 here, but it is not necessary. The _387 floating
; point routines in this file do not have any 387 specific op-codes,
; only 387 specific optimizations. (And plus my MS QuickAssembler does not
; recognize the .386/.387 directives.) -Wes
;
.286
.287
start_387:
cmp fractype,JULIAFP ; julia or mandelbrot set?
je short dojulia_387 ; julia set - go there
; Mandelbrot _387 initialization of stack
dec cx ; always requires at least 1 iteration
; the fpu stack is shown below
; st(0) ... st(7)
; b (already on stack)
fld inity ; Cy b
fld initx ; Cx Cy b
fld st(1) ; Cy Cx Cy b
fadd parmy ; Py+Cy Cx Cy b
fld st ; Py+Cy Py+Cy Cx Cy b
fmul st,st ; (Py+Cy)^2 Py+Cy Cx Cy b
fld st(2) ; Cx (Py+Cy)^2 Py+Cy Cx Cy b
fadd parmx ; Px+Cx (Py+Cy)^2 Py+Cy Cx Cy b
fmul st(2),st ; Px+Cx (Py+Cy)^2 (Py+Cy)(Px+Cx) Cx Cy b
fmul st,st ; (Px+Cx)^2 (Py+Cy)^2 (Py+Cy)(Px+Cx) Cx Cy b
; which is the next x^2 y^2 xy Cx Cy b
jmp short top_of_cx_loop_387 ; branch around the julia switch
dojulia_387:
; Julia 387 initialization of stack
; note that init and parm are "reversed"
; b (already on stack)
fld parmy ; Cy b
fld parmx ; Cx Cy b
fld inity ; y Cx Cy b
fld st ; y y Cx Cy b
fmul st,st ; y^2 y Cx Cy b
fld initx ; x y^2 y Cx Cy b
fmul st(2),st ; x y^2 xy Cx Cy b
fmul st,st ; x^2 y^2 xy Cx Cy b
top_of_cx_loop_387: ; x^2 y^2 xy Cx Cy b
fsubr ; x^2-y^2 xy Cx Cy b
fadd st,st(2) ; x^2-y^2+Cx xy Cx Cy b
fxch ; xy x^2-y^2+Cx Cx Cy b
; FADD is faster than FSCALE for 387
fadd st,st ; 2xy x^2-y^2+Cx Cx Cy b
fadd st,st(3) ; 2xy+Cy x^2-y^2+Cx Cx Cy b
; now same as the new
; y x Cx Cy b
cmp outside,-2 ; real, imag, mult, or sum ?
jg no_save_new_xy_387 ; if not, then skip this
fld st(1) ; x y x Cx Cy b
fstp newx ; y x Cx Cy b
fst newy ; y x Cx Cy b
no_save_new_xy_387:
cmp inside,-100 ; epsilon cross ?
jne end_epsilon_cross_387
call near ptr epsilon_cross ; y x Cx Cy b
jcxz pop_stack_387 ; if cx=0, pop stack
end_epsilon_cross_387:
cmp cx,oldcolor ; if cx >= oldcolor
jae end_periodicity_check_387 ; don't check periodicity
call near ptr periodicity_check_287_387 ; y x Cx Cy b
jcxz pop_stack_387 ; if cx=0, pop stack
end_periodicity_check_387:
cmp show_orbit,0 ; is show_orbit clear
je no_show_orbit_387 ; if so then skip
call near ptr show_orbit_xy ; y x Cx Cy b
no_show_orbit_387:
; y x Cx Cy b
fld st(1) ; x y x Cx Cy b
fld st(1) ; y x y x Cx Cy b
fmul st(3),st ; y x y xy Cx Cy b
fmulp st(2),st ; x y^2 xy Cx Cy b
fmul st,st ; x^2 y^2 xy Cx Cy b
fld st ; x^2 x^2 y^2 xy Cx Cy b
fadd st,st(2) ; x^2+y^2 x^2 y^2 xy Cx Cy b
cmp potflag,0 ; check for potential
je no_potflag_387
fst magnitude ; if so, save magnitude
no_potflag_387:
fcomp st(6) ; x^2 y^2 xy Cx Cy b
fstsw ax
sahf
ja over_bailout_387
;less than or equal to bailout
loop top_of_cx_loop_387 ; x^2 y^2 xy Cx Cy b
; reached maxit, inside
mov oldcolor,0FFFFh ; check periodicity immediately next time
mov ax,maxit
sub kbdcount,ax ; adjust the keyboard count
mov realcolor,ax ; save unadjusted realcolor
mov ax,inside_color
cmp inside,-59 ; zmag ?
jne no_zmag_387
fadd st,st(1) ; x^2+y^2 y^2 xy Cx Cy b
fimul maxit ; maxit*|z^2| x^2 y^2 xy Cx Cy b
; When type casting floating point variables to integers in C, the decimal
; is truncated. When using FIST in asm, the value is rounded. The following
; line cause the positive value to be truncated.
fsub round_down_half
fist tmp_word ; tmp_word = |z^2|*maxit
fwait
mov ax,tmp_word
shr ax,1 ; |z^2|*maxit/2
inc ax ; |z^2|*maxit/2+1
no_zmag_387:
pop_stack_387:
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -